Towards Scalable Dataset Construction: An Active Learning Approach

By Brendan M. Collins, Jia Deng, K. Li, Li Fei-Fei. 2008

146 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Optimization & Training

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

In Optimization & Training

Citations in the corpus

LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop (2015, background): “In many ways, this process is similar to active learning [14, 18, 3, 20].”
AutoML: A Survey of the State-of-the-Art (2019, methodology): “For example, the active learning method [66] selects the most “uncertain” unlabeled individual examples for labeling by a human, and then iteratively labels the remaining data.”
The Unreasonable Effectiveness of Noisy Data for Fine-Grained Recognition (2015, methodology): “h for collecting large quantities of ne-grained data. Active learning and other human-in-theloop systems have previously been used to create datasets in a more cost-ecient way than manual annotation [74,12,47], and our goal is to compare this more traditional approach with simply using noisy data, p”
What Do I Annotate Next? An Empirical Study of Active Learning for Action Localization (2018, background): “computer vision tasks, such as image classification [4,22,25,36,37,53] and object detection [64], or to construct large-scale image and video datasets [16,66,72].”
iCoseg: Interactive co-segmentation with intelligent scribble guidance (2010, methodology): “In computer vision, active learning has been used for object categorization [15], classifying videos [28], ranking images by informativeness [27] and creating large datasets [9].”
Learning by Asking Questions (2017, background): “Common selection criteria include entropy [25], boosting the margin for classifiers [1, 12] and expected informativeness [20].”
Generating Diverse Image Datasets with Limited Labeling (2016, background): “A few semi-supervised approaches have been developed that minimize human effort for dataset creation using an active learning framework [35, 9, 7].”
How far can you get with a modern face recognition test set using only simple features? (2009, background): “However, there are signiﬁcant downsides to this approach as well.”
Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop (2015, background): “People have looked at this problem from an active learning perspective [11, 15, 2].”
Large Scale Visual Recognition (2012, background): “In addition, a closely related question is how to develop active learning techniques [145] to optimize the selection of images and the granularity of annotation.”
Unsupervised Detection of Regions of Interest Using Iterative Link Analysis (2009, background): “Its main limitation is that much of the previous work requires additional assumptions such as a small number of seed images in the beginning [13], texts and HTML tags associated with images [19], and user-labeled images [6].”
MOTS: Multi-Object Tracking and Segmentation (2019, background): “While this is somewhat similar to an active learning setting [11, 56], we leave the decision which objects to annotate with our human annotators to guarantee that all annotations reach the quality necessary for a long-term benchmark dataset (c.”
Well Begun Is Half Done: Generating High-Quality Seeds for Automatic Image Dataset Construction from Web (2014, methodology): “But previous works mainly focus on the second step, paying relative less attention to the first step – the seeds are either labeled by human [10,14,13,5] (high cost) or labeled according to top-ranked images [14,13] or noisy surrounding text [1,16] (low quality).”
From region similarity to category discovery (2011, background): “As a result, much research has been devoted to designing efficient schemes for collecting training data for supervised object recognition [2, 3, 24].”
Bootstrapping Fine-Grained Classifiers : Active Learning with a Crowd in the Loop (2013, methodology): “As previously mentioned, active learning has been used to create pedestrian and object classifiers [1] [5].”
OPTIMOL: Automatic Online Picture Collection via Incremental Model Learning (2007, unknown): “352 true “accordions” out of 1659 images). 3. Princeton-23 ( Collins et al. 2008 )”
Automatic Image Dataset Construction from Click-through Logs Using Deep Neural Network (2015, background): “[4] proposed to construct image dataset in a semiautomatic manner by using active learning.”
Max-margin additive classifiers for detection (2009, background): “This complements a range of dataset collection efforts from semi-automatic [2, 25, 34, 5, 31] with 10,000-600,000+ images to fully manual but unpaid [32] with 50,000+ labeled objects to more traditional datasets [9, 13].”